LS-W4-T5-SM-Emotions

Model Description

This is a fine-tuned version of the google/flan-t5-small model, trained for the specific task of emotion classification from text.
The model takes a text input and generates a single word indicating the primary emotion.
It was fine-tuned on the dair-ai/emotion dataset.

• Developer: Linkspreed x Web4 AI
• Base Model: google/flan-t5-small
• Model Type: Encoder-Decoder (Text-to-Text)

---

Intended Use

This model is intended for research and educational purposes.
It can be used to classify the sentiment of short texts, such as social media posts, comments, or short sentences, into one of six categories:

• joy
• sadness
• anger
• love
• fear
• surprise

---

Training Data

The model was fine-tuned on the dair-ai/emotion dataset, which contains 20,000 English social media messages.

• Training set: 16,000 examples
• Validation set: 2,000 examples
• Test set: 2,000 examples

⚠️ Note:
The training data is highly imbalanced, with joy and anger being the most frequent emotions.
This may lead to a bias where the model over-predicts these two classes and performs poorly on the less frequent ones.

---

Training Details

Training Hyperparameters

The following hyperparameters were used during training:

• learning_rate: 5e-05
• train_batch_size: 8
• eval_batch_size: 8
• seed: 42
• optimizer: AdamW (torch fused) with betas=(0.9, 0.999), epsilon=1e-08
• lr_scheduler_type: linear
• num_epochs: 3

Training Setup

• Framework: PyTorch & Hugging Face Transformers
• Hardware: NVIDIA T4 GPU

---

How to Use

You can use this model directly with the Hugging Face pipeline for quick inference:

from transformers import pipeline

model_id = "Web4/LS-W4-T5-SM-Emotions"
analyst = pipeline("text2text-generation", model=model_id)

text_to_analyze = "sentiment: I am so happy about my new job!"
result = analyst(text_to_analyze)
print(result)

# Example output:
# [{'generated_text': 'joy'}]